Development Considerations
Consider investment costs
Before facing development, it is necessary to understand the cost limitation that will be used by the software we are developing. We do not face the same costs in developing a security system that will be executed to handle security in a 3-hour music concert, as a system that will be operating 24 hours in a Bank.
How do calls to face detection and recognition services work? (if we do both in the cloud)
• Assume that we have a face detected in the image, we are talking about 3 transactions, 3 API calls.
- 1 face detected returns 1 face ID – 1 API call
- 1 identified face returns 1 face ID – 1 API call
- We get 1 person, return data corresponding to the individual – 1 API call
• Assume we have three faces detected in the image, we are talking about, 5 API calls.
- 3 faces detected returns 3 face ID (the limit of the number of faces detected depends on the company that provides the service). – 1 API call
- 3 faces identified returns 3 face ID – 1 API call
- I get the data of the 3 people – 3 API calls
How do we translate these calls to values?
In another post I will delve into the pros and cons of the different offers that exist in the market, let’s also see above the values that AWS vs Azure handles.
- As part of the free offer AWS.
- Free use limit: 12 months / 5000 images per month.
- Up to 1M calls per month – 1USD
- After 9M calls per month – 0.80USD
- Calls: Amazon Rekognition counts the actual number of images analyzed as the number of images processed. FAQ here.
- As part of the free offer Azure.
- Free use limit: None / 30,000 transactions free per month making up to 20 transactions per minute.
- Up to 1M calls per month – 1USD
- After 5M calls per month – 0.60USD
- Calls: For all other operations, each API call will be counted as a transaction. FAQ here.
Face detection and recognition
Once a image has been captured, my suggestion is to make an offline face detection, this will reduce the cost of calling a service for detection. There are some Open Source frameworks that are suggested for this task, such as OpenCv, (Open Source Computer Vision Library) is an open source computer vision and machine learning software library. O TensorFlow, un pre-trained and optimized model to identify hundreds of classes of objects, including people, activities, animals, plants, and places.
Once the face is detected, we proceed with the recognition. Another suggestion is that the image to be sent to the cloud for such processing is only that of the detected face. This implies that you crop the image locally, which will reduce latency.
Posible problems
- If we are facing the development to be used in facial recognition in videos, consider that the face detected is a face that is in motion. This will prevent people to detect cover their face with another image, that a face is detected in magazines in scenes or posters, or mobile devices.
- Consider how the selected service acts for recognition of faces covered with caps, glasses, scarves, haircuts, or facial beauty.
Some items regarding security
This is where the need for anti-spoofing solutions come into play. We rely on liveness detection to validate an individual’s identity. As mentioned before we need to verify whether an individual is actually present or using a photo to spoof the system. There are many Anti-Spoofing Techniques like eye blink detection, wearable detection and as mentioned before3D cameras are the most reliable means of anti-spoofing. Precise pixel depth information can provide high accuracy against presentation attacks because we can tell the difference between a face and a flat shape.
Conclusion
Face recognition is only the beginning of implementing AI to improve our security. A human face is just one of the objects to be detected. Other objects can be identified in the same manner. Moreover, it’s possible to develop a multimodal biometric system, combining two or more biometric credentials, for example, face and voice recognition.
Technology for sure has matured and is slowly settling into the lives of people due to its usability and has made lives easier. Right from its invention to its accessibility, artificial intelligence is here not just to stay, but to grow and conquer
Ivana Tilca 
Leave a Reply